Power Amplifiers: Classes A, B, AB, C & D

Delivering watts to loudspeakers, motors, antennas and transducers. Learn output-stage requirements (voltage swing, current, low output impedance), conduction angle and amplifier classes (A, B, AB, C, D, plus G/H), efficiency and device dissipation, crossover distortion and its cure with AB biasing (V_BE multiplier), emitter ballast resistors, complementary and quasi-complementary stages, Class-D PWM operation with output LC filters, protection (current limiting, SOA, thermal), and choosing a class for an application — with Python computing output power, efficiency and dissipation for classes A, B and D.

Small-signal amplifiers care about gain and noise; power amplifiers care about how much of the supply power reaches the load and how much becomes heat. Efficiency decides heatsink size, battery life and cost.

The class of an amplifier describes how long each output device conducts during a signal cycle — from all the time (Class A) to switching fully on and off (Class D).

An analogy

Class A is a car engine idling at full throttle with the brakes on, ready for anything but wasting fuel; Class B is two cyclists taking turns pushing, each resting half the time; Class D is a light switch flicked so fast that the average brightness follows the music.

Amplifier classes

Defined by conduction angle and operating mode:

Crossover distortion and biasing

In pure Class B, neither transistor conducts while |v_in| < V_BE, so the output has a flat step at zero crossings — highly audible at low volume. Class AB biases both devices at a small quiescent current (tens of mA) with a V_BE multiplier mounted on the heatsink so the bias tracks temperature. Emitter resistors (0.1–0.47 Ω) provide local feedback against thermal runaway and help paralleled devices share current. Global negative feedback further reduces residual distortion.

Class D

The input is compared with a high-frequency triangle (300 kHz–2 MHz) to create PWM; a MOSFET half- or full-bridge switches the supply; an LC low-pass filter recovers the audio. Losses are conduction (I²R_DS(on)), switching and gate drive. Challenges: EMI, dead-time distortion, supply pumping in half-bridges, and filter interaction with the speaker. Filterless designs rely on speaker inductance for short cables.

Protection and reliability

Power stages need current limiting (V_BE-sensing or SOA protection), thermal shutdown, DC-offset detection to protect speakers, soft-start/muting against turn-on thumps, and heatsinks sized for worst-case dissipation (Class B peaks near 0.64 of full swing; music has high crest factor so average power is lower). Reactive loudspeaker loads make voltage and current out of phase, stressing devices beyond resistive-load calculations — check SOA with phase angles up to ±60°.

Formulas

Worked example: Output power

Given: 24 V peak into 8 Ω

Find: P_o.

Answer: 36 W

Worked example: Class-B heat

Given: ±24 V into 4 Ω

Find: Worst-case dissipation per device.

Answer: ≈ 14.6 W per device

Worked example: Maximum class-B efficiency

Given: V_p = V_CC

Find: η.

Answer: 78.5%

Class-AB complementary output stage

          +VCC
            │
  driver ── Q1 (NPN)
    │   ┌── V_BE multiplier (≈ 2·V_BE, thermally coupled)
    │   │       R_E 0.22 Ω
    └───┤         ├────── speaker 8 Ω ── GND
        │       R_E 0.22 Ω
        └── Q2 (PNP)
            │
          −VCC

Two emitter followers share the load: Q1 sources current on positive half-cycles, Q2 sinks on negative ones. A V_BE multiplier biases both slightly on (Class AB) to remove the dead zone around zero (crossover distortion); small emitter resistors stabilise the bias current against thermal runaway.

Efficiency and dissipation by class

import math
# Power amplifier classes: efficiency and device dissipation for an 8 ohm load on +/-20 V rails
VCC, RL = 20.0, 8.0
Vp_max = VCC - 2.0                               # ~2 V lost in drivers/emitter resistors
for frac in (1.0, 0.5, 0.1):
    Vp = Vp_max * frac
    Po = Vp**2 / (2 * RL)
    Ps_B = 2 * VCC * Vp / (math.pi * RL)         # class B, split supply
    print(f"Vp = {Vp:5.1f} V: Po = {Po:6.2f} W, class-B supply {Ps_B:6.2f} W, eff {Po/Ps_B*100:5.1f}%, loss/device {(Ps_B-Po)/2:5.2f} W")
Vp_worst = 2 * VCC / math.pi
Pd_max = (2 * VCC**2 / (math.pi**2 * RL)) / 2
print(f"class-B worst-case dissipation at Vp = 2VCC/pi = {Vp_worst:.1f} V: {Pd_max:.2f} W per device")
# Class A (single-ended, current-source load) for the same 18 V peak: bias IQ >= Vp/RL
IQ = Vp_max / RL
print(f"class A: IQ = {IQ:.2f} A, idle dissipation {2*VCC*IQ:.0f} W for {Vp_max**2/(2*RL):.2f} W out ({Vp_max**2/(2*RL)/(2*VCC*IQ)*100:.1f}%)")
# Class D: RDS(on) 50 mOhm, plus ~1 W switching and driver loss
Io = Vp_max / RL / math.sqrt(2)
Pcond = Io**2 * 0.05                             # half-bridge: one switch always conducts
print(f"class D at full power: conduction {Pcond:.2f} W, efficiency ~ {Vp_max**2/(2*RL)/(Vp_max**2/(2*RL)+Pcond+1.0)*100:.0f}% (1 W switching+driver)")
# Vp =  18.0 V: Po =  20.25 W, class-B supply  28.65 W, eff  70.7%, loss/device  4.20 W
# Vp =   9.0 V: Po =   5.06 W, class-B supply  14.32 W, eff  35.3%, loss/device  4.63 W
# Vp =   1.8 V: Po =   0.20 W, class-B supply   2.86 W, eff   7.1%, loss/device  1.33 W
# class-B worst-case dissipation at Vp = 2VCC/pi = 12.7 V: 5.07 W per device
# class A: IQ = 2.25 A, idle dissipation 90 W for 20.25 W out (22.5%)
# class D at full power: conduction 0.13 W, efficiency ~ 95% (1 W switching+driver)

Class B reaches 71% at full swing but only 7% at low levels, and each device's dissipation peaks (5.1 W) at about 64% of full swing — so the heatsink is sized for that point, not full power. Class A needs 90 W from the supply for a 20 W output. Class D keeps ~95% efficiency, which is why it dominates portable and high-power audio.

Common mistakes

Applications

Key takeaways

Going further